The first half of this talk will focus on using unsupervised machine learning for the purpose of wave analysis. The available magnetic field data from the terrestrial magnetosphere, solar wind and planetary magnetospheres exceeds over 1 million hours. Identifying plasma waves in these large data sets is a time consuming and tedious process. We propose a solution to this problem. We demonstrate how Self-Organizing Maps can be used for rapid data reduction and identification of plasma waves in large data sets. We use 72,000 fluxgate and 110,000 search coil magnetic field power spectra from the Magnetospheric Multiscale Mission and show how the Self-Organizing Map sorts the power spectra into groups based on their shape. Organizing the data in this way makes it very straightforward to identify power spectra with similar properties and therefore this technique greatly reduces the need for manual inspection of the data. We suggest that Self-Organizing Maps offer a time effective and robust technique, which can significantly accelerate the processing of magnetic field data and discovery of new wave forms.
The second half of the talk will focus on the analysis of velocity distribution functions (VDFs). The analysis of the thermal part of VDFs is fundamentally important for understanding the kinetic physics that governs the evolution and dynamics of space plasmas. However, calculating the proton core, beam, and alpha-particle parameters for large data sets of VDFs is a time-consuming and computationally demanding process that always requires supervision by a human expert. We developed a machine learning tool that can extract proton core, beam, and alpha-particle parameters using images (2D grid consisting pixel values) of VDFs. A database of synthetic VDFs was generated, which was used to train a convolutional neural network that infers bulk speed, thermal speed, and density for all three particle populations. We generated a separate test data set of synthetic VDFs that we used to compare and quantify the predictive power of the neural network and a fitting algorithm. The neural network achieves significantly smaller root-mean-square errors to infer proton core, beam, and alpha-particle parameters than a traditional fitting algorithm.